Fig. 1: Data representation of MD simulations.

a MD simulation of honeycomb structures allows us to predict compressive behavior. b Depending on the hierarchical architecture of the super-honeycomb, different stress behaviors result. c We subsequently encode different super-honeycomb lattices into vectors representing structure, and slice stress curves into progressive windows of stress evolution. In doing so, we train an ML model to predict stress curves given an initial structure and stress condition. d Once trained, the model can be deployed on a hierarchical honeycomb structure, parametrized by the cell wall thicknesses of each hierarchical hexagonal ring, at an initial unstressed state to predict the full stress-strain behavior.